Currently, the workflow_zero_shot_object_detection requires a GPU with CUDA. Thus, automated tests on GitHub Actions cannot be run, as there only a CPU is provided on the default runner. Supporting a CPU for inference would enable to include the test into the regularly performed tests.